A framework to predict the price of energy for the end-users with applications to monetary and energy policies

The study asks how to rigorously quantify and predict the average price of energy faced by end-use consumers in the United States and assess how it responds to major events and policy changes. The authors introduce the Energy Price Index (EPIC), a comprehensive measure aggregating prices across all energy feedstocks and end-use sectors, weighted by actual end-use demand. EPIC aims to fill the gap left by sector-specific price measures and financial indices that do not reflect end-user energy product prices. The work emphasizes the importance of a unified, predictive index for policy design and evaluation, given the sensitivity of energy markets to technological, monetary, and geopolitical shifts.

Existing energy-related financial indices (e.g., S&P 500 Energy Index, MSCI US IMI Energy 25/50, S&P GSCI Energy) are driven by market capitalization or world production weights and focus largely on company equities or commodity futures, often limited to oil and gas, and do not capture the full spectrum of end-user energy prices across sectors. Prior forecasting studies typically focus on individual energy commodities or sectors (electricity, natural gas, crude oil, petroleum products) and use shorter forecasting horizons. Thus, the literature lacks a comprehensive, demand-weighted, price index across all energy feedstocks and end-use sectors with demonstrated predictive capability for present and future periods. EPIC addresses these gaps by integrating both supply-demand mechanisms and prices across the entire energy landscape and by introducing a rolling horizon forecasting methodology to estimate contemporaneous and future demand weights.

EPIC construction and data: The EPIC index represents the average price of energy ($/MMBtu) to U.S. end-use sectors (residential, commercial, industrial, transportation). The framework identifies 56 energy products directed to end-use sectors, originating from crude oil, natural gas, coal, nuclear, solar, wind, hydroelectric, geothermal, and biomass. Energy flows into the intermediate electric power sector are excluded to avoid double counting; only electricity delivered to end-use sectors is included. Monthly consumption (in energy units) and prices (in $/energy unit) for each product are obtained from EIA and other public sources. For each month m and product p, the demand weight is w_mp = D_mp / Σ_p D_mp. EPIC is computed as EPIC_m = Σ_p (w_mp * c_mp), with c_mp the product price. Rolling horizon forecasting: Because end-use demand weights are released with a 2–3 month lag and for policy analysis forecasts are needed, a rolling horizon model is developed. The method trains each calendar month separately to capture strong seasonality in energy demand. It uses the previous three years of monthly observations (36 months) to predict current and future month weights. Four approaches (two weight-based, two demand-based) were tested over lookback windows of 24, 36, 48, and 60 months. The selected method (Approach 2: weight-based) minimizes the sum of squared errors between predicted and historical product weight vectors for the target month, with a 36-month lookback, as it produced the lowest errors among tested options. Prediction propagates forward in a rolling manner: stage T uses T−3, T−2, T−1; stage T+1 uses T−2, T−1, T; etc., enabling up to four years ahead by iteratively using predicted weights. Policy simulations: - Crude oil tax case: Assumptions include crude oil heating content 5.721 MMBtu/bbl, refinery efficiency 90%, negligible petroleum to power generation, and inelastic short-run crude oil demand. Historical EPIC adjustments apply the tax to petroleum product shares; future impacts (to 2024) use predicted petroleum product weights and EIA Annual Energy Outlook 2020 Reference projections for total demand and nominal shares. - Renewable energy targets/subsidies case: The policy affects only the electric power sector. For each non-fossil feedstock (nuclear, hydro, wind, biomass, geothermal, solar), target weights within electricity are imposed and remaining electricity feedstock weights are normalized to sum to one. Levelized costs come from Lazard (2008–2013) and EIA AEO (2014–2020). Changes in EPIC from shifting electricity mix (ΔEPIC_1) and from tax credits/subsidies (ΔEPIC_2 = w_elec * Tax_credit/1e6) are computed for past and future (to 2024) using predicted electricity product weights and AEO projections.

EPIC series and forecasting accuracy: - EPIC was computed monthly from Jan 2003 to Jun 2020 (values shown in the paper’s Fig. 2). - Rolling horizon prediction performance over 174 months (Jan 2006–Jun 2020) shows very low errors. The square roots of average sums of squared errors are approximately: 1st year 1.8808%, 2nd year 2.0874%, 3rd year 2.2329%, 4th year 2.3641%. The maximum monthly error’s square root is 8.1764% over the test period, indicating strong predictive capability even when using only predicted values in later years. Policy case study 1: Crude oil tax (historical impacts Jan 2003–Jun 2020): - A $10.25/bbl tax increases EPIC by $1.019/MMBtu (5.60%). - A $25/bbl tax increases EPIC by $2.484/MMBtu (13.66%). - Total revenue over the historical period ranges from $302.886 billion ($2.5/bbl) to $3,028.862 billion ($25/bbl). Average annual revenue scales about $17.308 billion per $2.5/bbl; for $10.25/bbl it is $70.962 billion/year. - Household impact example (2015): Baseline annual household energy use 77.1 MMBtu with EPIC $18.01/MMBtu implies $1,389.06/year. A $2.5/bbl tax would have increased EPIC by $0.2432/MMBtu (1.35%), raising annual costs by $18.76 to $1,407.82. A $10.25/bbl tax would have raised costs by $76.90 (5.54%) to $1,465.95. Policy case study 1 (future, Jul 2020–Jun 2024): - Predicted average EPIC increases: $10.25/bbl → +$0.977/MMBtu; $25/bbl → +$2.384/MMBtu. - Revenue: approximately $147.882 billion in total per $5/bbl increase over four years. Policy case study 2: Renewable energy targets/subsidies (historical impacts at maximum target weights): - Without subsidy (0 $/MMBtu), EPIC average % change at max target: nuclear +0.118%, hydro −0.602%, biomass −0.026%, geothermal −0.032%, solar +0.257%, wind −0.929%. - With $9/MMBtu subsidy, EPIC declines at max target by: nuclear −2.549% (avg annual budget $34,080M), hydro −2.024% ($18,176M), biomass −0.160% ($1,704M), geothermal −0.094% ($795M), solar −0.187% ($5,680M), wind −2.085% ($14,768M). Policy case study 2 (future, Jul 2020–Jun 2024, at max targets): - With 0 $/MMBtu subsidy: nuclear +0.657%, hydro −0.198%, biomass +0.054%, geothermal −0.019%, solar −0.090%, wind −0.143%. - With $9/MMBtu subsidy: nuclear −2.107% (budget $38,004M), hydro −1.672% ($18,017M), biomass −0.084% ($1,900M), geothermal −0.083% ($887M), solar −0.551% ($6,334M), wind −1.341% ($16,468M). - Example: wind at 13% share reduces EPIC by 0.143% with no subsidy; at 13% with $9/MMBtu subsidy, EPIC decreases by 1.341% with about $16.5B annual budget. Overall, hydro and wind increases in the electricity mix tend to reduce EPIC even without subsidies; nuclear and biomass require moderate subsidies to reduce EPIC on average.

The EPIC framework provides a comprehensive, demand-weighted measure of the average energy price to U.S. end-users, integrating all major feedstocks and end-use sectors. By accurately forecasting demand weights via a rolling horizon method that captures strong monthly seasonality, the framework can estimate current and future EPIC values despite data lags. This enables quantitative policy analysis: crude oil taxation scenarios translate directly into measurable increases in EPIC and predictable revenue streams, while renewable electricity targets and subsidies reveal trade-offs between budgetary costs and reductions in overall energy prices. The findings indicate that, under plausible targets, expanding hydro and wind shares can lower the average energy price even without subsidies, whereas nuclear and biomass generally require subsidies to yield price reductions. EPIC’s strong predictive performance and holistic scope make it a practical tool for policymakers to design, compare, and optimize energy and monetary policies with respect to consumer energy prices.

The paper introduces EPIC, a novel, predictive index of the average price of energy to end-use consumers in $/MMBtu, spanning the full U.S. energy landscape. It contributes: (1) a comprehensive formulation combining end-use demand weights and product prices across all major feedstocks; (2) a rolling horizon forecasting approach that overcomes data lags and accurately predicts current and future demand weights; and (3) policy applications demonstrating how EPIC quantifies impacts of crude oil taxes and renewable electricity targets/subsidies, including associated revenues and subsidy budgets. Results show that hydro and wind expansion can reduce EPIC without subsidies, while crude oil taxes materially increase EPIC with substantial revenue potential. Future work will extend EPIC to evaluate responses to financial/monetary shocks and technological advances, jointly optimize multiple energy sources under production targets and constraints, incorporate up-to-date technology characterizations for levelized costs, and leverage AI methods for multi-horizon forecasting, with the objective of informing federal renewable energy policy that mitigates climate change while maintaining energy affordability.

The analysis relies on publicly available data (primarily EIA) with a 2–3 month reporting lag; the rolling horizon forecasting mitigates but does not eliminate this limitation. Policy simulations incorporate specific assumptions: crude oil demand is inelastic in the short run; petroleum use for electricity generation is negligible; refinery efficiency is set at 90%; crude oil heating content is fixed at 5.721 MMBtu/bbl. Renewable electricity case studies assume targets are attainable with existing resources and current production costs, affect only the electric power sector (other sector demands unchanged), and use levelized cost estimates from Lazard (2008–2013) and EIA AEO (2014–2020). Forecasted effects to 2024 depend on predicted weights and AEO Reference projections for future demand and nominal shares. These assumptions bound generalizability and imply that results may vary under different elasticity estimates, technology costs, or sectoral interactions not modeled here.